LETTER TO THE EDITOR
Cell Research (2017) 27:147-150. www.nature.com/cr
Structural basis for blocking PD-1-mediated immune suppression by therapeutic antibody pembrolizumab Cell Research (2017) 27:147-150. doi:10.1038/cr.2016.77; published online 21 June 2016
Dear Editor, PD-1 is a type I immune inhibitory transmembrane receptor of the CD28 family that modulates the activity of T cells in peripheral tissues [1]. It is expressed in T cells, B cells, monocytes, natural killer cells and many tumor-infiltrating lymphocytes [2]. Binding of PD-1 to its ligands PD-L1 and PD-L2 reduces T-cell activity [3]. Thereby, under normal conditions, the interaction of PD-1 with PD-L1 or PD-L2 prevents excessive lymphocyte activation and maintains immune tolerance to self-antigens by negatively regulating the immune response [3]. However, PD-L1 is often overexpressed in different tumors including lymphoma, melanoma, nonsmall-cell lung cancer and other types of cancer [2]. As a result, tumor cells attenuate T-cell signaling to evade immune surveillance [4]. Blocking PD-1/PD-L1 interaction has been shown to restore T-cell activation and antitumor response, providing the rationale for therapeutic intervention using PD-1/PD-L1 as target [5]. Currently two monoclonal antibody-based drugs targeting PD-1 are in clinical trials. One is nivolumab or Opdivo from Bristol-Myers Squibb. The other is pembrolizumab or Keytruda, a therapeutic IgG4 antibody developed by Merck. Crystal structures of mouse PD-1 (mPD-1) in complex with human PD-L1 (hPD-L1), mPD-1 complexed with mouse PD-L2 (mPD-L2) and human PD-1 (hPD-1) in complex with hPD-L1 have revealed the structural basis of PD-1’s interaction with its ligands [6-8]. Crystal structure of the full-length pembrolizumab was also reported recently [9]. However, how pembrolizumab specifically recognizes hPD-1 is still unknown. Herein, we report the crystal structure of pembrolizumab Fab (antigen-binding fragment) in complex with hPD-1, revealing the molecular basis for the blockade of hPD-1/hPD-L1 interaction by pembrolizumab. Crystal structure of the hPD-1/pembrolizumab Fab complex (hPD-1/Fab) was determined at a resolution of 2.9 Å (Supplementary information, Table S1). hPD-1 and pembrolizumab Fab form a 1:1 complex (Figure 1A), consistent with the stoichiometry determined by previous
results [10]. hPD-1 is made up of a canonical β-sandwich immunoglobulin variable (IgV) topology with a disulfide bond between Cys54 and Cys123. Structural comparison of hPD-1 with apo-hPD-1 (PDB: 3RRQ) and hPD-1 structure extracted from the hPD-1/hPD-L1 complex (PDB: 4ZQK) shows that hPD-1 in the hPD-1/Fab complex resembles the conformation observed in the hPD-1/ hPD-L1 complex. The pembrolizumab Fab in the complex exhibits a canonical β-sandwich immunoglobulin fold closely resembling the full-length pembrolizumab antibody (Supplementary information, Figure S1) [9]. The interaction of PD-1 with pembrolizumab Fab buries ~1 774 Å2 surface area, and the hPD-1/Fab interface can be divided into two sub-interfaces. Sub-interface I mainly encompasses the C′D loop of hPD-1 and pembrolizumab Fab’s complementary determining regions (CDRs) L1, L3, H2 and four β-strands of framework region (FR), which interact through polar, charged and hydrophobic contacts (Figure 1B). The most notable feature of this sub-interface is that the C′D loop of hPD1 protrudes into a groove formed by the CDRs and FR of pembrolizumab Fab. Specifically, Asp85 of hPD-1 establishes a salt bridge with ArgH99 of FR (hereafter residues of the Fab light chain and heavy chain are designated by superscript chain identifiers L and H, respectively). The side chain of Ser87 forms hydrogen bond with ArgH99 of FR. Interestingly, two arginines Arg86 and ArgL96 are involved in a T-shaped stacking interaction. The backbone of C′D loop residues Glu84, Ser87, Gln88 and Gly90 are held in place by hydrogen bonds with side chains of TyrL36, TyrH35, AsnH59 and ThrH58, respectively. Furthermore, Pro89 of hPD-1 inserts into a cavity formed by side chains of TyrH33, TyrH35, AsnH52 and AsnH59 of β-stands 1, 2 and 3, and the main chains of GlyH50, IleH51, GlyH57 and ThrH58 of β-stands 1 and 2. Sub-interface II is dominated by hydrophilic interactions and brings together residues in the C, C′ and F strands of hPD-1 and CDRs L1 and H3 of Fab (Figure 1C). The side chains of Asn66 and Lys78 of hPD-1 form hydrogen bonds with the backbone groups of ArgH102 and TyrH101, respectively. The side chain of Thr76 of hPD-1 is
npg
Figure 1 Structural basis for the blockade of hPD-1/hPD-L1 interaction by pembrolizumab. (A) Overall structure of the hPD-1/pembrolizumab Fab complex. hPD-1 is shown in light blue, and the light and heavy chains of Fab are in wheat and pale green, respectively. The CDR loops and the β-strands of pembrolizumab that are involved in interactions are labeled. (B) View of sub-interface I in hPD-1/pembrolizumab Fab complex. Residues involved in the interaction are shown as sticks and labeled. Hydrogen bonds are shown in dash lines. (C) View of sub-interface II in hPD-1/pembrolizumab Fab complex. (D) Sequence alignment of the C′D loop in ectodomains of PD-1. Secondary structural elements of hPD-1 are shown on top of the alignment while those of mPD-1 are shown at the bottom. (E) ELISA data showing the binding of pembrolizumab to hPD-1 or hPD-1 mutants, and mPD-1. (F) Superposition of the hPD-1/pembrolizumab Fab complex with hPD-1/hPD-L1. hPD-L1 is shown in magenta. For simplicity, only hPD-1 in hPD-1/pembrolizumab Fab is shown in light blue.
npg Structural basis for PD-1-mediated immune suppression
148
SPRINGER NATURE | Cell Research | Vol 27 No 1 | January 2017
Zhenkun Na et al. npg 149
hydrogen bonded to the side chain of TyrH101. In addition, PheH103 of CDR H3 inserts into a hydrophobic pocket formed by Val64 and Pro83 of hPD-1 and TyrL34 of CDR L1. The extensive interactions at the hPD-1/Fab interface are consistent with the high binding affinity of pembrolizumab to the hPD-1, with an apparent disassociation constant (K D) of 27 pM. In the previously published hPD-1 structures (PDB: 3RRQ and 4ZQK), the C′D loop of hPD-1 is disordered and is assumed to be highly flexible [8, 10]. However, the C′D loop of hPD-1 in our hPD1/Fab complex structure is well ordered, as evidenced by its well-defined electron density, and it contributes to sub-interface I with Fab. Although mPD-1 and hPD1 share 60% sequence identity and an IgV topology, hPD-1 lacks the additional C″ strand observed in mPD1 (Figure 1D). Moreover, Asp85 and Arg86 in hPD-1 are substituted by Gly85 and Leu86 in mPD-1, respectively. Mutations in hPD-1, D85G and R86L show significant differences in their binding affinities with pembrolizumab. D85G abolishes hPD-1 binding to pembrolizumab as determined by ELISA (Figure 1E). This can be attributed to the disruption of the salt bridge with ArgH99 of Fab, which can conceivably impair the PD-1/Fab complex assembly. However, R86L did not affect the binding affinity between hPD-1 and pembrolizumab. These results are consistent with earlier data showing that pembrolizumab displays low binding affinity toward mPD-1 (Patent: WC500190992). Structural superposition of the hPD-1/pembrolizumab Fab complex and the hPD-1/hPD-L1 complex shows that pembrolizumab Fab and hPD-L1 interact with hPD1 through overlapping surface regions, suggesting that pembrolizumab and hPD-L1 can exclude each other from binding to hPD-1 (Figure 1F). Although the C′D loop in the sub-interface I contributes predominantly to the binding affinity of permbrolizumab, the C′D loop is disordered in the hPD-1/hPD-L1 complex, suggesting that it is not important for the hPD-1/hPD-L1 interaction. Consistent with this observation, the overlapping regions are mainly located in the sub-interface II, where the antigen-binding site of permbrolizumab Fab largely overlaps with the regions of hPD-L1 that interact with hPD-1. A second ligand for PD-1 is PD-L2, which shares 34% sequence identity with PD-L1 and exhibits 3-fold higher binding affinity for PD-1 [7]. Given that mPD-L2 and hPD-L2 share a sequence identity of 72%, we modeled the hPD-1/hPD-L2 complex based on the structures of hPD-1/hPD-L1 and mPD-1/mPD-L2. Structural superposition of hPD-1/pembrolizumab Fab complex and modeled hPD-1/hPD-L2 complex suggest that pembrolizumab Fab would also compete with hPD-L2 for binding to www.cell-research.com | Cell Research | SPRINGER NATURE
hPD-1 (Supplementary information, Figure S2) through overlapping regions similar to those observed between hPD-1/pembrolizumab Fab and hPD-1/hPD-L1. Taken together, these observations suggest a mechanism by which pembrolizumab outcompetes PD-L1 or PD-L2 for binding to hPD-1. In summary, we have reported the crystal structure of the pembrolizumab Fab in complex with the ectodomain of hPD-1. Pembrolizumab Fab uses its CDRs and FR to interact with the C′D loop of hPD-1, which appears unstructured in previously published reports. The epitope consists of several discontinuous segments of hPD-1, which overlap with the region that interacts with hPD-L1 or hPD-L2, suggesting a mechanism by which pembrolizumab prevents the binding of hPD-L1 or hPD-L2 to hPD-1. These results have implications for the design and improvement of mAb drugs targeting hPD-1. The atomic coordinates and structure factors for hPD1/pembrolizumab Fab complex structure have been deposited into Protein Data Bank under the accession code of 5JXE. Additional details of the methods are described in Supplementary information, Data S1.
Acknowledgments We would like to thank the beamline scientists at the European Synchrotron Radiation Facility in France for assistance of X-ray data collection. This work was supported by the Agency for Science, Technology and Research in Singapore.
Zhenkun Na1, Siok Ping Yeo2, Sakshibeedu R Bharath1, Matthew W Bowler4, 5, Esra Balıkçı1, Cheng-I Wang2, Haiwei Song1, 3 1
Institute of Molecular and Cell Biology, 61 Biopolis Drive, Singapore 138673, Singapore; 2Singapore Immunology Network, 8a Biomedical Grove, Singapore 138648, Singapore; 3Department of Biochemistry, National University of Singapore, 14 Science Drive, Singapore 117543, Singapore; 4European Molecular Biology Laboratory, Grenoble Outstation, 71 avenue des Martyrs, CS 90181 F-38042 Grenoble, France; 5Unit of Virus Host-Cell Interactions, Univ Grenoble Alpes-EMBL-CNRS, 71 avenue des Martyrs, CS 90181 F-38042 Grenoble, France Correspondence: Haiwei Songa, Cheng-I Wangb a Tel: +65 65869700 E-mail: [email protected] b Tel: +65 64070083 E-mail: [email protected]
References 1 Pardoll DM. Nat Rev Cancer 2012; 12:252-264. 2 Okazaki T, Chikuma S, Iwai Y, et al. Nat Immunol 2013; 14:12121218. 3 Hamid O, Robert C, David A, et al. N Engl J Med 2013; 369:134-144. 4 Rivoltini L, Carrabba M, Huber V, et al. Immunol Rev 2002; 188:97113.
npg Structural basis for PD-1-mediated immune suppression
150 5 Riley JL. N Engl J Med 2013; 369:187-189. 6 Lin DY, Tanaka Y, Iwasaki M, et al. Proc Natl Acad Sci USA 2008, 105:3011-3016. 7 Lazar-Molnar E, Yan Q, Cao E, et al. Proc Natl Acad Sci USA 2008; 105:10483-10488. 8 Zak KM, Kitel R, Przetocka S, et al. Structure 2015; 23:2341-2348. 9 Scapin G, Yang X, Prosise WW, et al. Nat Struct Mol Biol 2015; 22:953-958. 10 Cheng X, Veverka V, Radhakrishnan A, et al. J Biol Chem 2013; 288:11771-11785.
on the Cell Research website.)
(Supplementary information is linked to the online version of the paper
© The Author(s) 2016
This license allows readers to copy, distribute and transmit the Contribution as long as it attributed back to the author. Readers are permitted to alter, transform or build upon the Contribution as long as the resulting work is then distributed under this is a similar license. Readers are not permitted to use the Contribution for commercial purposes. Please read the full license for further details at - http://creativecommons.org/ licenses/by-nc-sa/4.0/
SPRINGER NATURE | Cell Research | Vol 27 No 1 | January 2017
Cell Research (2017) 27:147-150. www.nature.com/cr
Structural basis for blocking PD-1-mediated immune suppression by therapeutic antibody pembrolizumab Cell Research (2017) 27:147-150. doi:10.1038/cr.2016.77; published online 21 June 2016
Dear Editor, PD-1 is a type I immune inhibitory transmembrane receptor of the CD28 family that modulates the activity of T cells in peripheral tissues [1]. It is expressed in T cells, B cells, monocytes, natural killer cells and many tumor-infiltrating lymphocytes [2]. Binding of PD-1 to its ligands PD-L1 and PD-L2 reduces T-cell activity [3]. Thereby, under normal conditions, the interaction of PD-1 with PD-L1 or PD-L2 prevents excessive lymphocyte activation and maintains immune tolerance to self-antigens by negatively regulating the immune response [3]. However, PD-L1 is often overexpressed in different tumors including lymphoma, melanoma, nonsmall-cell lung cancer and other types of cancer [2]. As a result, tumor cells attenuate T-cell signaling to evade immune surveillance [4]. Blocking PD-1/PD-L1 interaction has been shown to restore T-cell activation and antitumor response, providing the rationale for therapeutic intervention using PD-1/PD-L1 as target [5]. Currently two monoclonal antibody-based drugs targeting PD-1 are in clinical trials. One is nivolumab or Opdivo from Bristol-Myers Squibb. The other is pembrolizumab or Keytruda, a therapeutic IgG4 antibody developed by Merck. Crystal structures of mouse PD-1 (mPD-1) in complex with human PD-L1 (hPD-L1), mPD-1 complexed with mouse PD-L2 (mPD-L2) and human PD-1 (hPD-1) in complex with hPD-L1 have revealed the structural basis of PD-1’s interaction with its ligands [6-8]. Crystal structure of the full-length pembrolizumab was also reported recently [9]. However, how pembrolizumab specifically recognizes hPD-1 is still unknown. Herein, we report the crystal structure of pembrolizumab Fab (antigen-binding fragment) in complex with hPD-1, revealing the molecular basis for the blockade of hPD-1/hPD-L1 interaction by pembrolizumab. Crystal structure of the hPD-1/pembrolizumab Fab complex (hPD-1/Fab) was determined at a resolution of 2.9 Å (Supplementary information, Table S1). hPD-1 and pembrolizumab Fab form a 1:1 complex (Figure 1A), consistent with the stoichiometry determined by previous
results [10]. hPD-1 is made up of a canonical β-sandwich immunoglobulin variable (IgV) topology with a disulfide bond between Cys54 and Cys123. Structural comparison of hPD-1 with apo-hPD-1 (PDB: 3RRQ) and hPD-1 structure extracted from the hPD-1/hPD-L1 complex (PDB: 4ZQK) shows that hPD-1 in the hPD-1/Fab complex resembles the conformation observed in the hPD-1/ hPD-L1 complex. The pembrolizumab Fab in the complex exhibits a canonical β-sandwich immunoglobulin fold closely resembling the full-length pembrolizumab antibody (Supplementary information, Figure S1) [9]. The interaction of PD-1 with pembrolizumab Fab buries ~1 774 Å2 surface area, and the hPD-1/Fab interface can be divided into two sub-interfaces. Sub-interface I mainly encompasses the C′D loop of hPD-1 and pembrolizumab Fab’s complementary determining regions (CDRs) L1, L3, H2 and four β-strands of framework region (FR), which interact through polar, charged and hydrophobic contacts (Figure 1B). The most notable feature of this sub-interface is that the C′D loop of hPD1 protrudes into a groove formed by the CDRs and FR of pembrolizumab Fab. Specifically, Asp85 of hPD-1 establishes a salt bridge with ArgH99 of FR (hereafter residues of the Fab light chain and heavy chain are designated by superscript chain identifiers L and H, respectively). The side chain of Ser87 forms hydrogen bond with ArgH99 of FR. Interestingly, two arginines Arg86 and ArgL96 are involved in a T-shaped stacking interaction. The backbone of C′D loop residues Glu84, Ser87, Gln88 and Gly90 are held in place by hydrogen bonds with side chains of TyrL36, TyrH35, AsnH59 and ThrH58, respectively. Furthermore, Pro89 of hPD-1 inserts into a cavity formed by side chains of TyrH33, TyrH35, AsnH52 and AsnH59 of β-stands 1, 2 and 3, and the main chains of GlyH50, IleH51, GlyH57 and ThrH58 of β-stands 1 and 2. Sub-interface II is dominated by hydrophilic interactions and brings together residues in the C, C′ and F strands of hPD-1 and CDRs L1 and H3 of Fab (Figure 1C). The side chains of Asn66 and Lys78 of hPD-1 form hydrogen bonds with the backbone groups of ArgH102 and TyrH101, respectively. The side chain of Thr76 of hPD-1 is
npg
Figure 1 Structural basis for the blockade of hPD-1/hPD-L1 interaction by pembrolizumab. (A) Overall structure of the hPD-1/pembrolizumab Fab complex. hPD-1 is shown in light blue, and the light and heavy chains of Fab are in wheat and pale green, respectively. The CDR loops and the β-strands of pembrolizumab that are involved in interactions are labeled. (B) View of sub-interface I in hPD-1/pembrolizumab Fab complex. Residues involved in the interaction are shown as sticks and labeled. Hydrogen bonds are shown in dash lines. (C) View of sub-interface II in hPD-1/pembrolizumab Fab complex. (D) Sequence alignment of the C′D loop in ectodomains of PD-1. Secondary structural elements of hPD-1 are shown on top of the alignment while those of mPD-1 are shown at the bottom. (E) ELISA data showing the binding of pembrolizumab to hPD-1 or hPD-1 mutants, and mPD-1. (F) Superposition of the hPD-1/pembrolizumab Fab complex with hPD-1/hPD-L1. hPD-L1 is shown in magenta. For simplicity, only hPD-1 in hPD-1/pembrolizumab Fab is shown in light blue.
npg Structural basis for PD-1-mediated immune suppression
148
SPRINGER NATURE | Cell Research | Vol 27 No 1 | January 2017
Zhenkun Na et al. npg 149
hydrogen bonded to the side chain of TyrH101. In addition, PheH103 of CDR H3 inserts into a hydrophobic pocket formed by Val64 and Pro83 of hPD-1 and TyrL34 of CDR L1. The extensive interactions at the hPD-1/Fab interface are consistent with the high binding affinity of pembrolizumab to the hPD-1, with an apparent disassociation constant (K D) of 27 pM. In the previously published hPD-1 structures (PDB: 3RRQ and 4ZQK), the C′D loop of hPD-1 is disordered and is assumed to be highly flexible [8, 10]. However, the C′D loop of hPD-1 in our hPD1/Fab complex structure is well ordered, as evidenced by its well-defined electron density, and it contributes to sub-interface I with Fab. Although mPD-1 and hPD1 share 60% sequence identity and an IgV topology, hPD-1 lacks the additional C″ strand observed in mPD1 (Figure 1D). Moreover, Asp85 and Arg86 in hPD-1 are substituted by Gly85 and Leu86 in mPD-1, respectively. Mutations in hPD-1, D85G and R86L show significant differences in their binding affinities with pembrolizumab. D85G abolishes hPD-1 binding to pembrolizumab as determined by ELISA (Figure 1E). This can be attributed to the disruption of the salt bridge with ArgH99 of Fab, which can conceivably impair the PD-1/Fab complex assembly. However, R86L did not affect the binding affinity between hPD-1 and pembrolizumab. These results are consistent with earlier data showing that pembrolizumab displays low binding affinity toward mPD-1 (Patent: WC500190992). Structural superposition of the hPD-1/pembrolizumab Fab complex and the hPD-1/hPD-L1 complex shows that pembrolizumab Fab and hPD-L1 interact with hPD1 through overlapping surface regions, suggesting that pembrolizumab and hPD-L1 can exclude each other from binding to hPD-1 (Figure 1F). Although the C′D loop in the sub-interface I contributes predominantly to the binding affinity of permbrolizumab, the C′D loop is disordered in the hPD-1/hPD-L1 complex, suggesting that it is not important for the hPD-1/hPD-L1 interaction. Consistent with this observation, the overlapping regions are mainly located in the sub-interface II, where the antigen-binding site of permbrolizumab Fab largely overlaps with the regions of hPD-L1 that interact with hPD-1. A second ligand for PD-1 is PD-L2, which shares 34% sequence identity with PD-L1 and exhibits 3-fold higher binding affinity for PD-1 [7]. Given that mPD-L2 and hPD-L2 share a sequence identity of 72%, we modeled the hPD-1/hPD-L2 complex based on the structures of hPD-1/hPD-L1 and mPD-1/mPD-L2. Structural superposition of hPD-1/pembrolizumab Fab complex and modeled hPD-1/hPD-L2 complex suggest that pembrolizumab Fab would also compete with hPD-L2 for binding to www.cell-research.com | Cell Research | SPRINGER NATURE
hPD-1 (Supplementary information, Figure S2) through overlapping regions similar to those observed between hPD-1/pembrolizumab Fab and hPD-1/hPD-L1. Taken together, these observations suggest a mechanism by which pembrolizumab outcompetes PD-L1 or PD-L2 for binding to hPD-1. In summary, we have reported the crystal structure of the pembrolizumab Fab in complex with the ectodomain of hPD-1. Pembrolizumab Fab uses its CDRs and FR to interact with the C′D loop of hPD-1, which appears unstructured in previously published reports. The epitope consists of several discontinuous segments of hPD-1, which overlap with the region that interacts with hPD-L1 or hPD-L2, suggesting a mechanism by which pembrolizumab prevents the binding of hPD-L1 or hPD-L2 to hPD-1. These results have implications for the design and improvement of mAb drugs targeting hPD-1. The atomic coordinates and structure factors for hPD1/pembrolizumab Fab complex structure have been deposited into Protein Data Bank under the accession code of 5JXE. Additional details of the methods are described in Supplementary information, Data S1.
Acknowledgments We would like to thank the beamline scientists at the European Synchrotron Radiation Facility in France for assistance of X-ray data collection. This work was supported by the Agency for Science, Technology and Research in Singapore.
Zhenkun Na1, Siok Ping Yeo2, Sakshibeedu R Bharath1, Matthew W Bowler4, 5, Esra Balıkçı1, Cheng-I Wang2, Haiwei Song1, 3 1
Institute of Molecular and Cell Biology, 61 Biopolis Drive, Singapore 138673, Singapore; 2Singapore Immunology Network, 8a Biomedical Grove, Singapore 138648, Singapore; 3Department of Biochemistry, National University of Singapore, 14 Science Drive, Singapore 117543, Singapore; 4European Molecular Biology Laboratory, Grenoble Outstation, 71 avenue des Martyrs, CS 90181 F-38042 Grenoble, France; 5Unit of Virus Host-Cell Interactions, Univ Grenoble Alpes-EMBL-CNRS, 71 avenue des Martyrs, CS 90181 F-38042 Grenoble, France Correspondence: Haiwei Songa, Cheng-I Wangb a Tel: +65 65869700 E-mail: [email protected] b Tel: +65 64070083 E-mail: [email protected]
References 1 Pardoll DM. Nat Rev Cancer 2012; 12:252-264. 2 Okazaki T, Chikuma S, Iwai Y, et al. Nat Immunol 2013; 14:12121218. 3 Hamid O, Robert C, David A, et al. N Engl J Med 2013; 369:134-144. 4 Rivoltini L, Carrabba M, Huber V, et al. Immunol Rev 2002; 188:97113.
npg Structural basis for PD-1-mediated immune suppression
150 5 Riley JL. N Engl J Med 2013; 369:187-189. 6 Lin DY, Tanaka Y, Iwasaki M, et al. Proc Natl Acad Sci USA 2008, 105:3011-3016. 7 Lazar-Molnar E, Yan Q, Cao E, et al. Proc Natl Acad Sci USA 2008; 105:10483-10488. 8 Zak KM, Kitel R, Przetocka S, et al. Structure 2015; 23:2341-2348. 9 Scapin G, Yang X, Prosise WW, et al. Nat Struct Mol Biol 2015; 22:953-958. 10 Cheng X, Veverka V, Radhakrishnan A, et al. J Biol Chem 2013; 288:11771-11785.
on the Cell Research website.)
(Supplementary information is linked to the online version of the paper
© The Author(s) 2016
This license allows readers to copy, distribute and transmit the Contribution as long as it attributed back to the author. Readers are permitted to alter, transform or build upon the Contribution as long as the resulting work is then distributed under this is a similar license. Readers are not permitted to use the Contribution for commercial purposes. Please read the full license for further details at - http://creativecommons.org/ licenses/by-nc-sa/4.0/
SPRINGER NATURE | Cell Research | Vol 27 No 1 | January 2017
